LMEB: Long-horizon Memory Embedding Benchmark

TL;DR

The paper introduces LMEB, a comprehensive benchmark with diverse datasets and tasks to evaluate long-horizon memory embeddings, highlighting the challenges and current limitations of models in complex memory retrieval scenarios.

Contribution

It presents LMEB, the first extensive benchmark for evaluating long-horizon memory embeddings across multiple memory types and tasks, filling a critical gap in the field.

Findings

Larger models do not always outperform smaller ones.

LMEB and MTEB measure different capabilities.

Traditional passage retrieval performance does not guarantee success in long-horizon memory retrieval.

Abstract

Memory embeddings are crucial for memory-augmented systems, such as OpenClaw, but their evaluation is underexplored in current text embedding benchmarks, which narrowly focus on traditional passage retrieval and fail to assess models' ability to handle long-horizon memory retrieval tasks involving fragmented, context-dependent, and temporally distant information. To address this gap, we introduce the Long-horizon Memory Embedding Benchmark (LMEB), a comprehensive framework for evaluating embedding models on complex, long-horizon memory retrieval. LMEB comprises 22 datasets and 193 zero-shot retrieval tasks spanning four memory types: episodic, dialogue, semantic, and procedural. These memory types differ in terms of level of abstraction and temporal dependency, capturing distinct aspects of memory retrieval that reflect the diverse challenges of the real world. We evaluate 15 widely used embedding models, ranging from hundreds of millions to ten billion parameters. The results reveal that (1) LMEB provides a reasonable level of difficulty; (2) Larger models do not always perform better; (3) LMEB and MTEB measure orthogonal capabilities. This suggests that the field has yet to converge on a universal model capable of excelling across all memory retrieval tasks, and that strong performance on traditional passage retrieval does not necessarily transfer to long-horizon memory retrieval. LMEB provides a standardized and reproducible framework that fills a key gap in memory embedding evaluation and supports future advances in long-term, context-dependent retrieval. LMEB is available at https://kalm-embedding.github.io/LMEB.github.io/.